Hi all.
Been trying for some time to get the Voice PE to output its TTS to an external media player.
I use a Windows PC (oldschool HTPC setup) in my living room with a good sound system and wanted the output here instead of the crapy builtin speaker.
Before this i had zero knowledge about ESPHome but i have been running HA instance for 4-5 years.
I took alot of inspiration from this thread Redirect Voice PE Replies to Sonos - Community Guides - Home Assistant Community.
ChatGPT helped me all the way here with the code.
Firstly i installed HASS.Agent on my PC(called sofa) and got it working.
Then i installed ESPHome and imported the Voice PE.
Voice PE → Play Replies on an External Media Player (No Double Audio)
Goal
Build a Voice PE satellite that:
Listens locally (microphone, wake word, LEDs all work normally)
Runs the full Assist pipeline in Home Assistant
Plays TTS replies on an external media player (e.g. media_player.sofa)
Does not speak locally at the same time
Works reliably, without race conditions or ESPHome YAML errors
In this setup, media_player.sofa is a Windows PC running Hass.Agent, exposed to Home Assistant as a media player.
The Robust Solution (Recommended)
Key Idea
Let Voice PE keep generating TTS, but:
- Mute the local Voice PE speaker
- Capture the generated TTS URL
- Hand off playback to Home Assistant
- Let HA play the reply on any media player (here: a Windows PC via Hass.Agent)
This is done using:
- an
input_texthelper as a bridge - a Home Assistant script for playback
- a small ESPHome override
ESPHome (Voice PE override)
What this does
- Ducks the local mixer
- Mutes the local Voice PE speaker
- Saves the TTS URL into Home Assistant
- Triggers the HA playback script
- Restores everything afterward
ESPHome YAML (override only)
substitutions:
name: home-assistant-voice-095a6b
friendly_name: Home Assistant Voice 095a6b
packages:
Nabu Casa.Home Assistant Voice PE: github://esphome/home-assistant-voice-pe/home-assistant-voice.yaml
esphome:
name: ${name}
name_add_mac_suffix: false
friendly_name: ${friendly_name}
api:
encryption:
key: ******************
wifi:
ssid: !secret wifi_ssid
password: !secret wifi_password
# ------------------------------------------------------------
# Voice PE reply redirect (NO double audio + no "blip" after idle)
#
# Home Assistant prerequisites:
# 1) Helper: input_text.voice_pe_tts_url
# 2) Script: script.voice_pe_play_reply_on_sofa
# (reads input_text.voice_pe_tts_url and plays it on media_player.sofa)
# ------------------------------------------------------------
voice_assistant:
# PRE-MUTE EARLY to prevent the "local starts speaking briefly then stops" issue after idle
on_intent_progress:
- if:
condition:
lambda: 'return !x.empty();'
then:
- logger.log:
level: DEBUG
format: "Redirect: pre-muting local Voice PE speaker before TTS begins."
- media_player.volume_set:
id: external_media_player
volume: 0.0
# Duck hard during TTS and ensure local speaker stays muted
on_tts_start:
- logger.log:
level: INFO
format: "Redirect: ducking local mixer + muting Voice PE speaker."
- mixer_speaker.apply_ducking:
id: media_mixing_input
decibel_reduction: 51
duration: 0s
- media_player.volume_set:
id: external_media_player
volume: 0.0
# Save the TTS proxy URL and trigger HA playback on sofa (Windows PC via Hass.Agent)
on_tts_end:
- logger.log:
level: INFO
format: "Redirect: saving TTS URL to HA helper + starting sofa playback script."
- homeassistant.service:
service: input_text.set_value
data:
entity_id: input_text.voice_pe_tts_url
value: !lambda |-
return x;
- homeassistant.service:
service: script.turn_on
data:
entity_id: script.voice_pe_play_reply_on_sofa
# Restore state when pipeline is fully finished
on_end:
- wait_until:
not:
voice_assistant.is_running:
- mixer_speaker.apply_ducking:
id: media_mixing_input
decibel_reduction: 0
duration: 0s
- media_player.volume_set:
id: external_media_player
volume: 1.0
I added this to script.yaml:
voice_pe_play_reply_on_sofa:
alias: Voice PE – Play reply on sofa
mode: restart
sequence:
- variables:
url: '{{ states(''input_text.voice_pe_tts_url'') }}'
- condition: template
value_template: '{{ url.startswith(''http'') }}'
- target:
entity_id: media_player.sofa
data:
media_content_id: '{{ url }}'
media_content_type: music
action: media_player.play_media
Added this to configuration.yaml:
input_text:
voice_pe_tts_url:
name: Voice PE last TTS URL
max: 255
Why This Works (and Why Others Fail)
No unsupported ESPHome YAML- No direct media_player hijacking
- No timing race conditions
- Works with any HA media player:
- Sonos
- Music Assistant
- Chromecast
- Windows PC via Hass.Agent
- Voice PE remains fully functional as a satellite
Voice PE still believes it is playing locally — but it’s muted — while Home Assistant takes over actual playback.
Result
You end up with a clean, professional Voice Assistant setup:
- One device listens
- Another device speaks
- No echo
- No hacks
- No flakiness
This is effectively how commercial multi-room assistants work — just implemented with full local control.
If you want, this approach can easily be extended to:
- restore exact previous volume
- room-aware replies
- multi-room announcements
- Music Assistant ducking
- LED sync with external playback
But as-is, this is already a production-grade solution.
I skipped the HASS.Agent part. Just needed my Home Assistant Voice PE to stop stuttering when replying (because it made me crazy). And now it TTS replies on my Google Nest Mini, or whatever media_player.* i want, with perfect flow and sound. Yess, I love it, VERY NAJS!!! 